GCAL: Adapting Graph Models to Evolving Domain Shifts

TL;DR

This paper introduces GCAL, a continual learning method for graph models that adapt to evolving domain shifts, effectively handling multiple out-of-distribution graphs and mitigating catastrophic forgetting.

Contribution

GCAL employs a bilevel optimization strategy with information maximization and variational memory generation, advancing graph domain adaptation for dynamic, real-world scenarios.

Findings

GCAL outperforms existing methods in adaptability.

GCAL effectively mitigates catastrophic forgetting.

GCAL demonstrates strong performance on evolving graph domains.

Abstract

This paper addresses the challenge of graph domain adaptation on evolving, multiple out-of-distribution (OOD) graphs. Conventional graph domain adaptation methods are confined to single-step adaptation, making them ineffective in handling continuous domain shifts and prone to catastrophic forgetting. This paper introduces the Graph Continual Adaptive Learning (GCAL) method, designed to enhance model sustainability and adaptability across various graph domains. GCAL employs a bilevel optimization strategy. The "adapt" phase uses an information maximization approach to fine-tune the model with new graph domains while re-adapting past memories to mitigate forgetting. Concurrently, the "generate memory" phase, guided by a theoretical lower bound derived from information bottleneck theory, involves a variational memory graph generation module to condense original graphs into memories. Extensive experimental evaluations demonstrate that GCAL substantially outperforms existing methods in terms of adaptability and knowledge retention.